Home // International Journal On Advances in Intelligent Systems, volume 15, numbers 3 and 4, 2022 // View article

Pinbone Detection of Japanese Shime-Saba by Near-Infrared Imaging

Authors:
Hisayoshi Ito
Oky Dicky Ardiansyah Prima
Takehiro Sasaki

Keywords: near-infrared imaging; bone detection; image geometry; convolutional neural network

Abstract:
There is a growing demand for new analytical techniques to manage the presence of fish bones more effectively. The detection of fish bones involves human inspection using touch and vision, which can lead to misjudgment. Many studies utilize X-ray machine vision approaches to effectively detect fish bones in quality control processes. However, this approach requires complex devices and carries the risk of radiation exposure. In this study, we attempted to detect the pinbone tip locations of shime-saba by using Near-Infrared (NIR) machine vision to highlight those features and quantify them based on image geometry and neural networks. The vinegar added to the shime-saba softens most of the bones, but the pinbones remain tough and need to be removed. Our approach is as follows. At first, the fish fillet is photographed with NIR transmitted through the fish fillet. The resulting image is correlated with the Gaussian template image. Quadratic surface equations are performed on the automatically defined Region of Interest (ROI) for this image and the convex-up shapes are selected as candidates for the pinbone tips. Rectangular areas (sub-images) of ten candidates are extracted and a Convolutional Neural Network (CNN) is constructed using these sub-images to determine the presence of pinbones. In the experiment, 95 samples of shime-saba fillets were captured in NIR, and 950 sub-images (225 contained pinbones) were extracted to train the CNN model. As a result, the CNN model was able to determine the bone with 84.9% accuracy.

Pages: 166 to 175

Copyright: Copyright (c) to authors, 2022. Used with permission.

Publication date: December 31, 2022

Published in: journal

ISSN: 1942-2679